Feature - Forecasting weather on the grid |
|
The Weather Research and Forecasting prognostic model (WRF) is an atmospheric simulation system that runs on parallel computing platforms. It is designed with the goal of being flexible, portable, and efficient. Within the field of meteorology, the Advanced Research WRF (ARW) is one of today's best-known weather research and forecasting models. ARW is suitable for use in a broad range of applications across scales ranging from meters to thousands of kilometers, including:
Numerical weather prediction models must use high resolution mesh grids in order to have sufficiently accurate and detailed results. Using high model resolution results in two natural consequences. The first is an increase in the number of computational grid points, necessary to keep the total domain size sufficiently large. The second is the reduction of the integration time step in order to avoid numerical instabilities. Both effects result in larger CPU and memory demands that ordinary computers cannot provide. The South-East region of Europe presents a particularly big challenge for meteorologists because of the complexity of the reproduction and forecasting of the airflow over complex terrain (e.g. mountain ranges, coastal area), such as the terrain in Croatia, Bosnia, and Herzegovina. |
In order to solve these computationally intense high resolution models, the SEE-GRID-SCI grid infrastructure was used. By deploying the model on the grid, we gained access to a large amount of processing power, memory, and storage capacity. The ARW model was deployed on the grid using generic gLite tools joined together in a few batch submission scripts, while the LHC Computing Grid File Catalogue was used for storing the model data. The results were quite positive. The gridified ARW has shown great improvements in storage capacity, the ability to run multiple model instances, and model execution speed (up to six times faster on 16 CPUs), running multiple. Today, the gridified ARW model is used for weather research purposes as well as operational forecasts. Most of the research focuses on wind simulations over coastal areas (e.g. bora winds over the North Adriatic) and inland (e.g. foehn wind effects over the city of Banja Luka). The model is also used regularly for operational weather prediction over Bosnia and Herzegovina. |